US 8200045 B2 Abstract The invention relates to an image processing method to generate, from a source image, an image of reduced size whose ratio between the width and the height is equal to a predetermined value, called reduced ratio. It comprises the following steps:
Claims(9) 1. An image processing method to generate, from a source image, an image of reduced size, called reduced image, whose ratio between the width and the height is equal to a predetermined value, called reduced ratio (RR), the method comprising:
selecting one rectangular image part in said source image,
extracting said rectangular image part to generate said reduced image, wherein, when the ratio between the width and height of said rectangular image part, called first ratio, is not equal to the reduced ratio, the width or the height of said rectangular image part is increased before the extracting by a first value on one first side of said image part and by a second value on one second side of said image part opposite to said first side, said first and second values being determined by:
calculating a first increase in width or height such that the ratio between said width and said height of said rectangular image part modified by said first increase is equal to said reduced ratio, calculating a first average saliency value from the values of perceptual interest associated with the pixels of the additional image part obtained by increasing the width or height of said rectangular image part of said first increase to said first side,
calculating a second average saliency value from the values of perceptual interest associated with the pixels of the additional image part obtained by increasing the width or height of said rectangular image part of said first increase to said second side,
calculating said first value as being the product of said first increase and the ratio between said first average saliency value and the sum of said first and second average saliency values, called total average saliency value, and
calculating said second value as being the product of said first increase and the ratio between said second average saliency value and said total average saliency value.
2. The method according to
3. The method according to
4. The method according to
5. The method according to
6. The method according to
7. The method according to
8. The method according to
9. The method according to
Description The invention relates to an image processing method to generate an image of reduced size from a source image. For navigation applications in a set of fixed images or videos, it is useful to be able to display all the images of videos at a single glance. For this purpose, it is interesting to generate a reduced size version of each fixed image or each image of the videos so that said images or videos are displayed simultaneously on a same screen and are able to be compared easily. Likewise, for a broadcast application of a video content on a mobile appliance having a small sized screen, e.g. on a mobile phone, it is necessary to generate a reduced size version of each image of the video in order to display them on the small sized screen. A method known to a person skilled the art to generate these reduced images or videos from a source image or video consists in subsampling said source image or video. In the case of a significant reduction in size some image parts are unusable by the user as they are too small. The purpose of the invention is to compensate for at least one disadvantage of the prior art. The invention relates to an image processing method to generate, from a source image, an image of reduced size whose ratio between the width and the height is equal to a predetermined value, called reduced ratio. It comprises the following steps:
According to a particular characteristic, if the first ratio is less than the reduced ratio, the width of the rectangular image part is increased by a first value to the right and by a second value to the left, the first and second values being determined according to the following steps:
According to another particular characteristic, if the first ratio is greater than the reduced ratio, the height of the rectangular image part is increased by a first value upward and by a second value downward, the first and second values being determined according to the following steps:
According to a particular embodiment, wherein the source image belongs to a sequence of several images, the method is applied to each image of the sequence and comprises, after the modification step of the width or height of the rectangular image part of a current image, a temporal adaptation step of the rectangular image part consisting in moving at least one of the sides of the rectangular image part if the side is separated from the corresponding side of the rectangular image part of the image that precedes the current image by a distance greater than a predetermined threshold such that the distance is reduced. According to another particular embodiment, wherein the source image belongs to a sequence of several images, the method is applied to each image of the sequence and comprises, after the modification step of the width or height of the rectangular image part of a current image, a temporal adaptation step of the rectangular image part consisting in moving all the sides of the rectangular image part if each side is separated from the corresponding side of the rectangular image part of the image that precedes the current image by a distance greater than a predetermined threshold such that the distances are reduced. According to another particular embodiment, wherein the source image belongs to a sequence of several images, the method is applied to each image of the sequence and comprises, after the modification step of the width or height of the rectangular image part of a current image, a temporal adaptation step of the rectangular image part consisting in modifying at least one parameter of the rectangular image part among the following parameters: the height, width and coordinates of the centre such that the modified parameter is equal to the median value of the set of values comprising the corresponding parameters of the rectangular image parts of the N images preceding the current image and the N′ images following the current image, with N and N′ positive integers. According to a particular characteristic, N=N′. According to another particular characteristic, the sequence being divided into groups of images called shots, the N images preceding the current image and the N′ images following the current image belong to the same shot as the current image. Advantageously, if the ratio between the width and height of the temporally adapted rectangular image part is not equal to the reduced ratio, the temporal adaptation step is followed by a modification step of the width or the height of the temporally adapted rectangular image part according to the values of perceptual interest associated with each pixel of the current source image such that the ratio between the width and the height of the temporally adapted and modified rectangular image part is equal to the reduced ratio. The invention will be better understood and illustrated by means of advantageous embodiments and implementations, by no means limiting, with reference to the figures attached in the appendix, wherein: The invention relates to a method for generating, from a source image, an image of reduced size whose ratio between the width and height is equal to a predetermined ratio, called reduced ratio and noted as RR. The invention described for an image can be applied to a sequence of images such as a video. In reference to Step 110 consists in calculating the ratio RF between the width PL and the height PH of the image part selected in step 100. Step 120 consists in comparing RF with RR:
In the same manner during a step 1312 and in reference to During a step 1313, an overall average saliency value SMT is calculated as follows: SMT=SMG+SMD. During a step 1314, the increase ALG in width to the left is calculated as follows:
Likewise, during a step 1315, the increase ALD in width to the right is calculated as follows:
The modified bounding box REM is shown in In reference to In the same manner during a step 1322 and in reference to During a step 1323, an overall average saliency value SMT is calculated as follows: SMT=SMB+SMH. During a step 1324, the increase AHH in height upwards is calculated as follows:
Likewise, during a step 1325, the increase AHB in height downward is calculated as follows:
The modified bounding box REM is shown in The step 140 consists in generating the reduced image by extracting, from the source image, the image part delimited by the modified bounding box REM. Advantageously according to the invention, the bounding box is modified to obtain a box whose ratio between the width and height is equal to the ratio RR such that the modified bounding box REM comprises pixels for which the perceptual interest is high. In the case where the bounding box RE determined in the step 100 is situated near one of the edges of the source image, it is possible that the additional boxes leave the source image. In this case, during the determination step of the modified bounding box REM, the pixels located outside of the source image are placed on the opposite side. According to a particular embodiment of the invention illustrated by The distances dd, dg, dh and db are modified in the following manner:
According to a variant, the distances dd, dg, dh and db are modified in the following manner:
THD is a predefined threshold. For a sequence of images in the CIF format, TDH=20. ddm, dgm, dhm and dbm are the modified values de dd, dg, dh and db respectively. α_{cv }is the temporal convergence speed. If α_{cv }is large then the position of the window AREM_{t }changes little with respect to the position of the window AREM_{t−1}. The sides of the modified bounding box REM_{t }are then moved in such a manner that the distance that separates the left side (respectively right side) of the temporally adapted modified bounding box AREM_{t }of the current image and the left side (respectively right side) of the temporally adapted modified bounding box AREM_{t−1 }of the previous image is equal to dgm (respectively ddm) and in such a manner that the distance that separates the top side (respectively bottom side) of the modified bounding box REM_{t }of the current image and the top side (respectively bottom side) of the temporally adapted modified bounding box AREM_{t−1 }is equal to dhm (respectively dbm). The temporal adaptation step 135 can modify the size of the bounding box. Hence, it may be necessary after this step to reapply the steps 110 to 130 so that the width to height ratio of the bounding box AREM_{t }is equal to RR. According to another embodiment of the invention, the temporal adaptation step 135 of the bounding box is carried out per shot. A shot is a group of consecutive images separated by two changes of shot of the other images of the sequence. For this purpose, parameters are extracted from the modified bounding boxes according to steps 110 to 130 in each image of a shot of the sequence. These parameters are the coordinates (Cx, Cy) of the centre of the modified bounding box REM, its height H and its width L. A temporal median filter of length N is applied to several images. It consists in sorting the values of each of the parameters according to an increasing order and in selecting the median value. In the case where the number of values is even, then the average of the two values of the middle is chosen. A calculation example is given for a filter of length N=9. The median is calculated for each parameter associated with the REM of the image of index t from the values of parameters associated with the 3 images preceding the image of index t and with the 3 images following the image of index t and listed in the table below:
The values of the parameter Cx classed in increasing order are the following: 227, 227, 230, 230, 231, 231, 233, 235 and 236. The median value is the value 231 that is then the new value of Cx for the temporally adapted bounding box of the image of index t (AREM_{t}). The median values for the other parameters are as follows:
In the image of index t, the bounding box after filtering is therefore the box centred on the pixel of coordinates (231, 172) of width 159 pixels and of height 131 pixels. The parameters thus modified for each bounding box are more correlated between each other, providing an increase in temporal coherence (or reducing the inter-image jerks). Naturally, for the N/2 first and N/2 last images of the shot, is it not possible to apply the median filtering as is. In this case, the median of the N first images of the shot (respectively the N last images of the shot) is applied. For example, if N=9, the parameters (Cx, Cy, L, H) of the bounding boxes of the 4 first images of the shot are the same. The value of each parameter is equal to the median value obtained by median filtering of the 9 first values of the corresponding parameter of the shot. Of course, the invention is not limited to the embodiment examples mentioned above. In particular, the person skilled in the art may apply any variant to the stated embodiments and combine them to benefit from their various advantages. The method described can be used for many applications such as the summary of a sequence of images, navigation, indexation of image sequences. The invention can also be used to improve the viewing of images or sequences of images on small screens such as PDA, mobile phone or digital camera screens. Patent Citations
Non-Patent Citations
Classifications
Legal Events
Rotate |